Shreyansh Lodha @staticvar_dev

30 | He/Him | Staff Engineer (Mobile) - @MutualMobile | Thoughts are my own 💜💛 staticvar.dev Hyderabad, India Joined March 2010

Tweets

1K
Followers

247
Following

681
Likes

39K

Shreyansh Lodha @staticvar_dev

3 days ago

@chrisbanes Silicon Valley 👀

0 0 0 180 0

View Details

Shreyansh Lodha @staticvar_dev

a week ago

I’ve used 2 things in tandem 1. I’ve setup lints for identifying some basic issues, ex: hardcoded numbers or strings. Or multiple classes in one file etc. 2. I’ve created 2-3 skill files which deal with different issues, one for writing better compose code, one for working across domains, another for writing useful tests only. I’ll agree the second option Isn’t the best and it alignes with the issue you mentioned where your skill instructions get compacted and the agents stop following them, its because every model and every harness works differently with each other. Although for me I’ve seen that if you use the right trigger words the skill gets invoked very well at least in codex with astra. If you want 100% reliability then agent hooks are your best bet but that would likely not scale well and shouldn’t be done.

1 0 0 23 0

View Details

Shreyansh Lodha @staticvar_dev

a week ago

Review for GPT 6 Astra, for Kotlin & Mobile development. I have been using it since it came out & have burnt roughly ~1B tokens. First off lets talk about the good parts 1. It pretty much always completes the task you give it. I've thrown a handful of different types of task at it & it has always done the task, except once where it said it did but it missed 1 out of 9 occurrences of the fix. (I can live with that) 2. Computer use is really good too, it can end to end test it's own code changes by spinning off emulator & simulator, navigating them & taking screenshots or verifying debug logs & reviewing them. (very autonomous) 3. It still sometimes write unnecessary tests, like GPT 5.6 Sol but its frequency is much less comparatively. 4. It does forget things sometimes, especially which were communicated earlier (like 10 prompts prior, probably due to compaction). And yes I do have the experimental context management flag enabled & I am using codex/chatgpt harness. 5. Much much much better to talk to & explains things without making you feel dumb & throwing a ton of jargons towards you (unlike 5.6 Sol) The issues that I ran into with Astra 1. One of my major issue with Astra was the amount of times it violated domain boundaries, it fails to understand which part of change should to go which layer. (Does koin injection inside a Composable, gives lifecycle state to view models) 2. This model requires a fair bit of steering and hand holding to get "good" code out of it. But can be fixed with skills. 3. The code it produces, I would not call it readable or pleasant to read & review. If you are working with a team which does care about how code looks & how things are built, this model would require a fair bit of custom "Skills" & a fair bit of steering. Which I believe can be offset by its capability of getting things done. If you or your team cares about just building & getting things done, I believe this model would do wonders for you. Especially with the level of autonomy this model provides. Astra is pretty pricy at API costs so be mindful how you use it & start with low reasoning, trust!

5 0 6 2K 2

View Details

Shreyansh Lodha @staticvar_dev

a week ago

@kashif_mehmood_ Yup, exactly And after getting tired of steering the model too many times, I just decided to write a skill to address it!

1 0 0 8 0

View Details

Shreyansh Lodha @staticvar_dev

a week ago

@infiloop2 Probably, had to create a skill to handle it but it worked well after that.

0 0 1 10 0

View Details

Shreyansh Lodha @staticvar_dev

a week ago

@mktpavlenko I understand that but it was something very minor so I’m okay with it. But if it was something important then I’d be much critical of it.

0 0 0 9 0

View Details

Shreyansh Lodha @staticvar_dev

a week ago

@ethereaglehq I couldn’t find any skill like that but I created something of my own to fix these kinds of problems.

1 0 0 34 0

View Details

Shreyansh Lodha @staticvar_dev

a week ago

Here are some images of the app I am working on, with Astra

0 0 0 59 0

View Details

etn. @etnshow

2 weeks ago

JUST IN: NVIDIA's $12,930,300,000 acquisition of Hugging Face contains an easter egg. The number 129,303 is the decimal conversion of Unicode point U+1F917. The 🤗 emoji.

119 747 14K 3.8M 2K

View Details

maria @maria_rcks

2 weeks ago

I WENT OUT TO BUY GROCERIES AND 3 MODELS DROPPED?

44 11 1K 60K 16

View Details

Shreyansh Lodha @staticvar_dev

3 weeks ago

And this is why we dont deserve nice things! This is really bad!

Binkg @dummerspast39

3 weeks ago

worst reasoning i have ever seen. if ur a plus peasant and want to escape the permanent underclass (unlimited codex by ABUSING chatgpt): github.com/totec448-spec/… - sub agents - auto compaction - goal - codex tools

20 3 212 48K 300

0 0 0 80 0

View Details

Shreyansh Lodha @staticvar_dev

3 weeks ago

@theo @atshrey While you are at fork please also add support for /side

0 0 0 102 0

View Details

Shreyansh Lodha @staticvar_dev

3 weeks ago

@ejmarchini @StunnerFPS That recovery was smoother than a slab of butter!

0 0 0 1K 0

View Details

Shreyansh Lodha @staticvar_dev

4 weeks ago

@ACTFibernet this is just ridicules, internet has been down for over 36 hours Totally unacceptable and your lack of accountability makes it worse! SR10181705229784

1 0 3 119 0

View Details

Raghav Aggarwal 👨‍💻 🚀 @avidRaghav

4 weeks ago

@Google recently announced AppFunctions in Google I/O which let's the an Android device's local AI agents to trigger and access your app's functionalities. So far we have focussed on making our apps usable to humans now it's time to make them usable for AI Agents. github.com/avidraghav/Age…

1 1 9 353 4

View Details

Shreyansh Lodha @staticvar_dev

4 weeks ago

@SisyphusPussy @grok Thats true! You do know me better!

0 0 0 8 0

View Details

Shreyansh Lodha @staticvar_dev

a month ago

@grok (dont embarrass me) Analyze my tweets, profile, and interactions and respond with the following: Age: Mental age: Hobbies: Personality disorder: IQ: Qualifying adjective to describe me: 5 favorite things:

Santosh Yadav @SantoshYadavDev

a month ago

Lets me try for fun @grok Analyze my tweets, profile, and interactions and respond with the following: Age: Mental age: Hobbies: Personality disorder: IQ: Qualifying adjective to describe me: 5 favorite things: